Frame approach to Persian verb generation for educational purposes
Constance Bobroff
Middle Eastern Studies
University of Texas at Austin
1 Univ Stn, F9400
Austin, TX 78712
[email protected]
Artem Lukanin
Faculty of Linguistics
South Ural State University
pr. Lenina 76, Chelyabinsk
Russian Federation, 454080
[email protected]
Abstract
This paper describes an on-line Persian
Verb Conjugator using a frame based
system. Created rules allow the generation
of Persian verb forms in conjunction with
a Past-Present stem lexicon to handle
exceptions. This program conjugates
prefixed and compound verbs as well as
independent verbs and creates full
paradigms of positive and negative verb
forms. In addition, the program is used to
generate interactive multiple-choice tests
for self-practice.
1
1.1
Input methods by the User
Most students will access PVC having been given
the URL of the PVC index of the most common
verbs
of
modern
standard
Persian
(http://persianintexas.org/courses/PRS506/PVClist.
html) listed in (Persian) alphabetical order along
with the English translation. The student may
Introduction
Along with dictionaries and grammars, many
languages have a book of the most common one
hundred or so verbs fully conjugated for all
persons and tenses, including the exceptions, for
students of that language. To date, the Persian
language lacks such a reference work. To fill this
void, the online Persian Verb Conjugator (PVC)
was created as a free, open-access pedagogical tool
for self-study or classroom use.
PVC is first and foremost a tool to teach the
mechanical conjugation of modern Persian verbs, a
most difficult task for beginning and intermediate
students of Persian.
The students should
understand that PVC, having been supplied with
the two stems of each verb, conjugates the verbs
dynamically, “on the fly” according to linguistic
rules with which it has been supplied and that they
too, instead of rote memorization can also perform
the same mechanical operations after learning the
same rules driving PVC.
Picture 1. Infinitives may be manually typed in
Perso-Arabic script or Latin transliteration from
any page on PVC.
proceed to the conjugation of any verb by clicking
either on the Perso-Arabic script version or the
transliterated version in Latin characters.
Other users typically come upon PVC having
typed or pasted some conjugated form of a verb
into a search engine such as Google hoping to
determine the infinitival form so that they may
then look up the meaning of the verb in a
dictionary.
The third way a user may call up the conjugation
of a verb is by manually entering the infinitive
from any page within PVC. The interface allows
the user to toggle between Perso-Arabic script or
Latin transliteration without having any need for
Persian script or extended Latin capabilities on the
user end.
1.2
Unicode Encoding Compliancy
Perso-Arabic: Unlike Arabic, Persian requires
Unicode (UTF-8) encoding for proper display of
all the letters of the alphabet. Because our target
audience consists of students, part of whose
training includes following scholarly and scientific
international standards, the common practice of the
substitution of Arabic (Microsoft Windows
codepage 1256) letters, specifically Yeh
(U+064A), Kaf (U+0643) and Space (U+0020) for
respectively, Persian Yeh (U+06CC), Kaf
(U+06A9) and ZWNJ (U+200C) in order that the
program be viewable in non-Unicode compliant
browsers and platforms was not an option.
Latin transliteration: The only character in the
extended Latin subset not commonly found on US
English keyboards is the “a with macron above”
(U+0101) to represent the long /a/ sound. PVC
allows the user to type “aa” (a common practice
among Persian users in everyday situations such as
email) which PVC then dynamically converts to
“ā”.
1.3
with Microsoft Office for Macs, Tahoma should be
disallowed for such educational web pages for Mac
users. Instead, the Geeza Pro font (unavailable on
Windows) is more suitable for Macs. Thus, the
PVC style sheet has been designed to display in
Tahoma for the Windows users and Geeza Pro for
Macs.
CSS Font-family and font-size
A few precautions were taken in consideration of
the beginning level learners of Persian. While
native speakers can read Persian fonts at very small
font sizes, the beginners require at least an 18 pt
font in order to not strain the eye. The well-hinted
Tahoma font is very readable even at low
resolution and is the font of choice for Windows.
On the other hand, owing to the faulty nature of the
Perso-Arabic subset of the Tahoma font supplied
Picture 2. Font display in Tahoma on a PC with
Internet Explorer.
Picture 3. Font display in Geeza Pro on a Mac with
Safari.
1.4
Color-coding system
The verb forms in Persian differ mainly in stems.
To help students to differentiate them a colorcoding system was developed. The verb forms
based on the Past stem are marked in yellow color,
the verb forms based on the Present stem are
marked in purple color. The stems are shown
before the paradigm serving mostly the purpose of
showing the legend of colors used in PVC. The
Perfect tenses are based on the Past Participle,
which in consequence is based on the Past stem.
To group these tenses together the verb forms of
the Perfect Subjective, Present Perfect and Past
Perfect are marked in green color. The same PastPresent stem color system is used in tests on
conjugation. The background of tests on Present
tenses (including the Imperative) is purple, while
the background on Past tenses is yellow. Such a
color-coding system provides users with an
additional clue in memorization of tenses in
Persian. (The self-test feature of PVC is taken up
below in section 2.4.)
1.5
Pedagogical Considerations
Persian is a language known for its literature,
especially poetry with ambiguities suggesting
limitless interpretations. It is a language which
rejects standardization at every level. Therefore,
each school of Persian has its own methods and
agreement among scholars and practitioners of the
language is rare.
Which verbs should be
considered in a learner’s basic verb list, tenses, the
names of the tenses, the six default pronouns, the
verb classification scheme, the transliteration
system and the spelling conventions all required
decision-making and compromise. In the end,
those methods which favor the needs of the
students, the target users and those which promote
scholarly conventions were chosen.
2
System Description
The Persian Verb Conjugator is an on-line
education service, which is written in PHP
programming language and is located at
http://www.laits.utexas.edu/persian/pvc/. It is
based on the frame approach, wherein every verb
form is a strictly defined frame. Each slot can
remain empty or filled with a pseudo-affix,
calculated differently for each verb. The pseudoaffix is a sequence of letters, which is not
isomorphous to the affix in traditional grammar.
For example, the Infinitive 'sukhtan' (‫ )ﺳﻮﺧﺘﻦ‬is
divided into the primordial participle 'sukh' and the
suffix 'tan' in (Fazel, 2006) whereas in PVC it is
divided into the Past stem 'sukht' and the Infinitive
pseudo-affix 'an'.
PVC generates all verb forms for the following
tenses, including positive and negative forms:
1) the Simple Past;
2) the Imperfect;
3) the Perfect Subjunctive;
4) the Past Progressive;
the Present Perfect;
the Past Perfect;
the Present Indicative;
the Present Progressive;
the Present Subjunctive;
the Future;
the Imperative.
There are no negative forms in the Progressive
tenses. For educational purposes it was decided to
omit verb forms in 1st and 3rd persons for the
Imperative in PVC. However, the missing persons
are channeled to special usage tests on the Optative
and Jussive grammatical functions. Thus PVC
generates up to 112 verb forms. Nevertheless, the
generation function used both in PVC and tests on
conjugation can generate up to 240 verb forms
(120 in Active voice and 120 in Passive voice).
This means there are 120 frames for generation of
these verb forms and 1 frame for generation of the
Past Participle needed in formation of passive
forms.
PVC takes infinitives in Perso-Arabic script or
transliteration and generates verb forms in the
corresponding script. In general PVC does not
require any lexicons to generate verb forms but it
uses an exceptions list (which we refer to as the
lexicon) with correspondences between the Past
stem and the Present stem for verbs not covered by
the PVC rules.
5)
6)
7)
8)
9)
10)
11)
2.1
Present Stem Derivation
Every Persian verb form is built upon one of two
different stems for surface forms. As the Past stem
is known before verbal paradigm generation, being
derived from the Infinitive, we have to derive the
Present stem from the Infinitive as well to generate
3 Present tenses and the Imperative. Though as
Megerdomian, (2004, p. 5) states, “the two stems
… cannot be derived from each other”, we
developed several rules for Present stem
derivation. These rules can be used to conjugate
verbs not present in lexicons, e.g. neologisms.
Due to the large number of exceptions, so far no
one has come up with a fool-proof scheme to
organize Persian verbs into patterns, however
Boyle (1966, pp. 30-36) has come up with a
manageable system which defines 10 patterns for
Present stem derivation. Boyle’s system defines a
regular pattern for each of the 10 classes and then
lists exceptions to each class. After analyzing these
rules we developed 10 rules for deriving Present
stems from the Infinitives written in transliteration
and 9 rules for verbs in Perso-Arabic script. The
difference is that we use only one rule for the
infinitives ending in -ndan (e.g. afkandan) and rdan (e.g. āvardan) instead of two. Thus, to get the
Present stem the final 3 letters must be deleted so
that the Present stems for these verbs will be
‘afkan’ and ‘āvar’. The verbs ending in these
pseudo-affixes,
but
requiring
additional
transformations (e.g. ‘bordan-bar’) are placed into
the lexicon of exceptions. The first step of the
algorithm is to search for the Infinitive in the
lexicon of exceptions. If the verb is not found, the
derivation rules are applied to get the Present stem
(see Picture 4). So, if the verb ‘bordan’ is not
placed into the lexicon of exceptions, we get a
wrong Present stem *bor.
Also we use two different rules for verbs ending
in -estan (e.g. bāyestan) and -stan (e.g. jastan),
while Boyle defines one rule for such verbs. To
derive the Present stems for these verbs the first
rule removes the pseudo-affix -estan, while the
second rule removes the pseudo-affix -stan and
adds -h, so that we get the correct stems ‘bāy’ and
‘jah’.
Inf
COMP = all
except last word
in lexicon?
yes
PresStem =
COMP+PresStem
no
compound?
yes
no
Present stem
derivation rules
generate verb
forms
Picture 4. PVC algorithm.
Thus the Present stem derivation algorithm is
defined so that the fourth letter from the end of the
Infinitive is compared with the list of letters.
Depending on every letter the definite number of
letters is subtracted from the end and a number of
certain letters are added to the subtracted form. In
some rules there is an additional condition
whereby we must look at the fifth letter from the
end to select a rule for deriving the Present stem.
Due to the fact that the short vowels are omitted
in words written in the Persian script, parsing of
infinitives in the Persian script is necessarily more
complex. But as we take the fourth letter from the
end of infinitives as the base for comparison, only
one short vowel can be in this position in
transliteration — letter ‘a’ (e.g. verb ‘zadan’). Thus
we have the same 10 rules for the Persian script,
but the ‘-adan’ rule is executed last. Here the base
of comparison is not the fourth, but the third letter
from the end as the second letter — the short
vowel ‘a’ — is omitted.
Difficulty still exists when we choose between
two rules — ‘-stan’ and ‘-estan’. Short vowel ‘e’ is
omitted and cannot be the base for comparison. As
we have only four examples for the ‘-stan’ rule we
have made a rule depending on their structure: two
of them consist only of four letters, i.e. there is one
letter before the pseudo-affix –stan (e.g. ‫)ﺟﺴﺘﻦ‬
while the fourth letter from the end of two others is
Alef (e.g. ‫)ﺧﻮاﺳﺘﻦ‬. Therefore the infinitives not
satisfying these conditions are supposed to be
verbs parsed by the ‘-estan’ rule.
Compounding is very productive in Persian,
even more productive than creation of new verbs
using –idan suffixing (about 90% of all Persian
verbs are compound). The most frequent verbs
used for the creation of compound verbs are kardan
(‫)ﮐ ﺮدن‬, shodan (‫)ﺷ ﺪن‬, dādan (‫)دادن‬, zadan (‫)زدن‬,
budan (‫)ﺑ ﻮدن‬, dāshtan (‫)داﺷ ﺘﻦ‬, sākhtan (‫)ﺳ ﺎﺧﺘﻦ‬,
etc. As the Present stems of most of them cannot
be derived using our rules (except for dāshtan and
sākhtan), they are listed in the lexicon of
exceptions. If such verbs are listed in the lexicon,
PVC can correctly produce the paradigms of the
compounds, based on them (see Picture 4). For
example, if verb 'kardan' is listed in the lexicon, the
Present stem of verb 'tabdil kardan' (‫ )ﺕﺒﺪﻳﻞ ﮐﺮدن‬will
be derived correctly (tabdil kon, ‫)ﺕﺒﺪﻳﻞ ﮐﻦ‬. The
Present stem for verb 'bāz dāshtan' (‫)ﺑﺎز داﺷﺘﻦ‬, for
example, will be derived correctly (bāz dār, ‫)ﺑﺎز دار‬
using only the –shtan rule (remove –shtan and add
–r) without the lexicon of exceptions.
Presently there are 95 exceptions in the lexicon.
As most of them are used for compounding, PVC
is estimated to correctly generate paradigms of 9095% of all Persian verbs (including possible
neologisms).
2.2
Verb Form Frames
Every verb form in PVC is a frame, i.e. a pattern,
where each slot is filled in with a pseudo-affix or a
non-conjugated verb of a compound verb. Each
slot can be empty, e.g. every tense frame has a
COMPOUND slot, which is filled in with the verb
attached to the main verb to form a compound
verb. If the given verb is compound, the
COMPOUND slot is filled in, otherwise it is left
empty (see verb ‘bar gozidan’ in Table 1).
Every pseudo-affix is calculated for every verb
before filling in the slots, e.g. if the final letter of
the Present stem is Alef, (a or ā), pseudo-affix
VyV is set to Persian Yeh, otherwise it is left
empty (see verb ‘afzudan’ in Table 1).
Additional rules can deny some frames. In the
current version of PVC there are only two verbs
denying frames: ‘budan’ does not have forms in
the Present Progressive, the Past Progressive, the
Perfect Subjunctive and the Past Perfect, denying
the corresponding frames; ‘dāshtan’ denies the
frames for the Present Progressive, the Past
Progressive and the Perfect Subjunctive.
The Passive voice in Persian is formed with the
Past Participle and the conjugated form of shodan
(‫ﺷﺪن‬, to become). Because this is a syntactical
construction and therefore does not involve any
morphological or phonological changes it was
decided not to generate passive forms in PVC
Slot 1
DAAR
dār
Slot 2
SG1
am
Slot 3
SPACE
SPACE
‫دار‬
‫م‬
dār
am
SPACE
SPACE
‫دار‬
‫م‬
SPACE
itself, but as with all matters relating to usage
rather than mechanical manipulation, to make an
additional self-test on Active/Passive voice
formation. It should be noted, that the base
function, used in both PVC and conjugation tests,
takes an additional parameter to return the required
frame in active or passive voice. When the
function is called with the passive voice parameter,
the Past Participle of the required verb is added to
the contents of the COMPOUND slot and the
PastStem and PresentStem variables are rewritten
with the Past and Present Stems of the verb shodan
(shod and shaw respectively). For example, if it is
required to generate the passive verb form for the
Past Progressive 3rd singular of ‘bar chidan’ ( ‫ﺑﺮ‬
‫)ﭼﻴﺪن‬, the contents of the COMPOUND slot ('bar ',
‫ )ﺑﺮ‬become 'bar chide ' (‫ )ﺑﺮ ﭼﻴﺪﻩ‬and then all the
required contents of the rest of the slots of the Past
Progressive 3rd singular frame are calculated based
on the verb ‘shodan’. The contents of the slots can
be seen in Table 2. So after concatenating slots the
resulting form will be 'dāsht bar chide mishod'
(‫)داﺷﺖ ﺑﺮ ﭼﻴﺪﻩ ﻣﯽﺷﺪ‬.
2.3
Control Panel
PVC was originally designed to be an open system.
That is, in order to encourage participation among
scholars and teachers of Persian, a facility was
added to allow users to submit new or modified
Slot 4
Slot 5
COMPOUND PREF_MI
bar+SPACE
mi
‫ﺑﺮ‬+SPACE
Slot 6
PresStem
gozin
Slot 7
VyV
‫ﻣﯽ‬+ZWNJ ‫ﮔﺰﻳﻦ‬
mi
afzā
‫ﻣﯽ‬+ZWNJ ‫اﻓﺰا‬
Slot 8
SG1
am
‫م‬
y
am
‫ی‬
‫م‬
Table 1. Present Progressive 1st singular frame with 2 examples: ‘bar gozidan’ (‫ )ﺑﺮ ﮔﺰﻳﺪن‬and
‘afzudan’ (‫)اﻓﺰودن‬.
Slot 1
Slot 2
Slot 3
DAASHT SPACE COMPOUND
dāsht
SPACE bar+SPACE+chide+SPACE
Slot 4
PREF_MI
mi
‫داﺷﺖ‬
‫ﻣﯽ‬+ZWNJ ‫ﺷﻮد‬
SPACE
‫ﺑﺮ‬+SPACE+‫ ﭼﻴﺪﻩ‬+SPACE
Slot 5
PastStem
shod
Table 2. Past Progressive 3rd singular frame with the Passive Voice parameter for verb ‘bar chidan’
(‫)ﺑﺮ ﭼﻴﺪن‬.
Picture 5. Control panel.
verbs or suggestions. However, after several
months, the only individuals taking advantage of
this facility appeared to be mischief-makers and so
this facility was disabled.
In the interests of future uses of PVC, it was
decided to use the lexicon of exceptions not only
for exceptions, but for providing additional
information about every verb:
• English translation;
• dialect;
• style;
• usage (if the verb is obsolete or not);
• transitiveness;
• change of state;
• Boyle pattern;
• level (basic or advanced);
• person for the passives.
The administrators has a special control panel,
where they can add new verbs, edit or tag existing
ones and delete incorrect verbs.
The same lexicon is also used as the base for
conjugation tests due to the fact that some
information is required only for test generation
purposes. For example, Boyle pattern number is
used to select verbs from the list belonging only to
one of ten patterns or to all of them to make a
comprehensive test on Boyle patterns. The level
field is used to generate conjugation tests for
students of different skill levels. The transitiveness
field is used to select only transitive verbs for the
test on Voice. Both 'tabdil kardan' (‫ )ﺕﺒﺪﻳﻞ ﮐﺮدن‬and
'tabdil shodan' (‫ )ﺕﺒﺪﻳﻞ ﺷﺪن‬verbs, for example, are
not selected for this test, because the former has
'only active' and the latter has 'only passive' values
in this field. This means that the program selects
only those verbs, which can be used in both active
and passive constructions.
Not all verbs can be used in all persons in
Passive voice. For example, verb 'bar chidan' ( ‫ﺑﺮ‬
‫ )ﭼﻴﺪن‬can be used only in 3rd person singular and
plural. That is why only these two verb forms are
used in tests on Passive voice for verbs with this
tag.
2.4
Conjugation Tests
However much PVC may be a useful tool in
performing mechanical operations on verb stems to
correctly conjugate verbs, from the new learner’s
perspective, the number of forms can be
overwhelming. The student may not be able to
digest and assimilate the data given in the form of
charts or paradigms such as presented in PVC.
Furthermore, PVC makes no claims to assist with
matters of usage. Therefore, a need was felt for an
interface between student and PVC whereby the
student could practice one kind of manipulation at
a time taking actual usage into consideration.
This interface has taken the form of interactive,
multiple-choice
tests
(http://laits.utexas.edu/persian/tests/index.php)
which are fed from PVC using the PVC algorithm.
The advantage of such tests in comparison to other
grammar tests is that they are dynamically created
every time meaning the student may retake them
multiple times and always have a fresh test
preventing monotony and fatigue. The randomly
chosen verbs are automatically conjugated into the
required verb forms using the verb form generation
function of PVC. The order of questions and
answers is randomized each time.
The tests take the verbs from the PVC lexicon
and thus require only the generation function. The
Present Stem Derivation algorithm is not used
here, because additional information is required for
generation of different tests (see section 2.3).
Presently the exceptions list consists of 182 verbs,
the true exceptions of which are 95. The rest of the
tagged verbs are listed there for the purposes of
generating the self-tests. Exception here refers to
any verb, the Present stem of which cannot be
derived from its Infinitive using our derivation
rules.
Some verbs are marked to be used in special
tests. Thus the verbs from Boyle patterns are
marked with the number of the pattern, e.g., Boyle
Pattern 1 test selects the verbs only from this
pattern. Other conjugation tests select verbs
randomly from the whole lexicon.
The generation function requires 6 arguments:
• the Past stem;
• the Present stem;
• the tense number;
• the form number (there are 6 forms: 3
persons singular and plural);
• the script (transliteration or the Persian
script);
• the voice (passive by default).
Every test is presented both in transliteration and
the Persian script allowing self-learning for
students of different levels. Many tests are
presented in two versions: basic vs. advanced
levels, the former being more suited to beginning
level students.
Different tests have different algorithms for
selecting questions and distractors (wrong answer
choices). For example, the test on Negative
Imperfect selects random questions of eight types
and wrong answers of four types. It is offered to
either translate a verb from English to Persian or to
change the verb form from the Present Indicative
to the Imperfect. Any Persian verb in Imperfect
can be translated by seven different sentences. For
example, the verb form 'midādand' can be
translated into English as 'They were in the habit of
regularly giving' or ‘They used to give’ or ‘They
would give’. Each Persian infinitive has one or
more English translations in the lexicon. A random
translation is taken from the lexicon to fill in the
required slot. The terminal slot can be a function
for generation of required the English verb form.
Three English-language functions were created to
generate Simple Present 3rd singular (e.g.
carry→carries), Participle I (e.g. tie→tying) and
Past forms (e.g. take→took, take→taken or
change→changed). The functions use special
phonological rules and a lexicon of irregular verbs
for this purpose. The wrong answers can be:
• the verb in different tense;
• the verb in different person;
• incorrectly spelled verb;
• a different verb in correct tense and
person.
It was noted that students of Persian often
mistakenly write the negative prefix separately
from the verb. While such practice is allowed in
transliteration, it is a true mistake in the Persian
script. For this reason, the wrong spelling ‫ﻧﻪ ﺧﻮردﻩ‬
‫ ﺑﺎﺷﻴﻢ‬can be used in this test as a distractor among
the multiple-choice answers along with the correct
spelling ‫ ﻧﺨﻮردﻩ ﺑﺎﺷﻴﻢ‬but not for the counterpart test
in the transliterated script.
2.5
Classroom experiences and feedback
PVC was integrated into the syllabus of the firstyear Persian class at the University of Texas at
Austin and was used by the students with great
success for the academic year 2006-2007. Since
the students generally have access to the internet,
either on their own laptops or through campus
computer labs, this online tool is always at hand.
However, the most important use of PVC in the
life of the students is in the self-testing
opportunities provided by the tests which not only
serve the student well in the initial teaching phase
but in review and maintenance as well. The
improved performance on class written tests and
speaking practice testifies to the effectiveness of
this tool. The only recommendation from the
students so far seems to be that they want more of
the same and more variety of tests.
2.6
Future Plans
So far, PVC has been primarily used for modern,
standard Persian. However, by tagging each verb,
there is scope for adding obsolete and dialectal
variants. There is also a need to add usage notes
for each verb and deal with suppletive forms and
verbs requiring cross referencing of various kinds.
All verbs should ideally be available in the
colloquial forms as well as the current written
styles.
The developers have kept all these
considerations in mind and the verbs are being
tagged accordingly with these future uses in mind.
It is hoped that one day, an audio component will
also be available for each form.
Acknowledgements
The online version of PVC received inspiration
and guidance from an earlier downloadable version
based on different technology and different
methodology and currently still available from its
creator, Ali Jahanshiri at
http://alijsh.googlepages.com/pvc.htm
The developers of the online PVC gratefully
acknowledge much help and guidance —
sometimes on a daily basis — from Dr. Gernot
Windfuhr, University of Michigan. We wish to
also thank University of Texas Liberal Arts
Instructional Technology Services for generously
providing server space including the facility for
long-distance collaboration involved in this project
and in particular we would like to mention Justin
Davis and Nick Lauland who keep the servers up
and running and always respond promptly to all
manner of emergencies.
References
John A. Boyle. Grammar of modern Persian.
Wiesbaden, Harrassowitz, 1966.
Navid Fazel. 2006. Academic Grammar of New
Persian.
http://www.fazel.de/dastur/EN/index.html
Karine Megerdomian. Finite-state morphological
analysis of Persian. In Proceedings of the
Workshop on Computational Approaches to
Arabic Script-based Languages. Coling 2004,
University of Geneva. August 28, 2004.